Week 3 of 6 · 2.5-3 hours
Image Recognition and Model Mistakes
See how an image classifier turns pixels into predictions, measure how well it works, and find and fix the mistakes it makes.
Big question: How does a classifier turn pixels into a prediction, and how do we find and fix the mistakes it makes?
By the end of this week
- Explain how an image classifier turns pixels into visual features, a prediction, and a confidence.
- Measure a classifier fairly using accuracy and category-level accuracy on a separate test set.
- Tell false positives from false negatives and spot the edge cases that cause them.
- Read a confusion matrix and propose targeted improvements to reduce a specific mistake.
Loading your progress…
- How an Image Classifier WorksFollow a photo from pixels to a label, and learn how an image classifier turns visual features into a prediction with a confidence score.
- Train and Test a ClassifierLearn why a classifier is checked on photos it never trained on, and measure how well it does using accuracy on a fair test set.
- Confuse and Improve the ModelFind the edge cases that fool a classifier, read its mistakes in a confusion matrix, and propose changes that would make it better.